gh-137944: use argparse instead of getopt in timeit
#137955
Conversation
Misc/NEWS.d/next/Library/2025-08-19-14-52-27.gh-issue-137944.gAFxTV.rst
Outdated
Show resolved
Hide resolved
| @@ -362,7 +365,6 @@ def format_time(dt): | |||
| % (number, 's' if number != 1 else '', | |||
| repeat, format_time(best))) | |||
|
|
|||
| best = min(timings) | |||
There was a problem hiding this comment.
FTR, best is already computed 5 lines above.
| scales = [(scale, unit) for unit, scale in units.items()] | ||
| scales.sort(reverse=True) |
There was a problem hiding this comment.
FTR, this is constant for each invokation, so I can move it outside the function definition.
|
Ok, there is one possible breaking change: $ python3.12 -m timeit '1+1' -n1 -r1
Traceback (most recent call last):
...
NameError: name 'n1' is not defined$ ./python -m timeit '1+1' -n1 -r1
1 loop, best of 1: 410 nsec per loopI need to find a way for argparse to gobble everything normally. I don't know if it's possible though. |
|
Answer: use argparse.REMAINDER |
There was a problem hiding this comment.
Ok, so there are a few bits that I could consider separate issues. For instance, if timeit.main() fails at the timer construction, then sys is not restored correctly.
On another note, scales could be extracted outside format_time function but that's mostly some cleanup code which doesn't really warrant a standalone PR.
So I can for now hold this PR because it's not really important. It's indeed a bit fragile and I'm worried about a possible unforseen change of behavior.
At some point, I wanted to reduce the cyclotomic complexity of
main(), but I ended up with more lines than before, which defeated the purpose. While the function is a bit long, it's not that hard to follow, so I would rather keep a smaller diff.argparseintimeit#137944